Closed Isratja45 closed 1 year ago
I think the OP is a bit quick but how cool would it be if we could automatically feed the Chatwoot conversation content into your own OpenAI through their API (at the cost of the user). Combine that with feeding extra content like manuals, guides, forums and even Discord chats into OpenAI, the resulting ChatGPT conversations could be a really wonderful opener before handing a conversation off to an agent.
I'm sure other chat services are working real hard on exactly this idea so it would be great if Chatwoot can be ahead of the curve here!
Thoughts?
Exploring ideas for some of the potential implementations of these language models/APIs into Chatwoot.The OpenAI language models provide the following capabilities.
Ref: https://platform.openai.com/docs/introduction/next-steps https://platform.openai.com/docs/models/overview
We could explore building numerous product features leveraging these capabilities.
ref: https://platform.openai.com/docs/guides/completion/inserting-text
These APIs provide us with the option to build various capabilities, sentiment analysis, translations, summarisation etc.
Reply Suggestions
Auto-populated reply suggestions for agents generated via the Conversation context, which can be sent by the agent within a couple of clicks. Examples: Gmail / Linkedin reply suggestionsAutocomplete Suggestions
Autocomplete suggestions while the agent is typing out the reply based on conversation context. examples: GmailSummarization
Summarise the large conversation into a gist so that agents can educate themselves about the conversation context in a single glance. Examples: Google Chat conversation summaryTransformation
Rephrase Conversations [Write different copies, Change the tone of text into helpful, professional etc.], Translate the text between languages etc.Bots
Bot support agents powered by ChatGPT/GPT3 APIs.Text Expansion
Write short notes or bullet points which will be expanded into full-fledged customer replies or support Center Articles. Examples: Notion AISentiment Analysis
System generated CSAT Ratings for conversations where the users haven't provided a feedback rating. This will help the administrators to improve the quality of support provided.ref: https://platform.openai.com/docs/guides/chat
Suited for Agent Bots
OpenAI now has a very specific model, gpt-3.5-turbo
which allows exploring the above-mentioned use cases with a Chat style API interface. examples use cases suggested by Open AI
This model is very cost-effective, but at the moment, it doesn't support
fine-tuning
.
ref: https://platform.openai.com/docs/guides/fine-tuning
GPT-3 has been pre-trained on a vast amount of text from the open internet. When given a prompt with just a few examples, it can often intuit what task you are trying to perform and generate a plausible completion. This is often called "few-shot learning."
Fine-tuning improves on few-shot learning by training on many more examples than can fit in the prompt, letting you achieve better results on a wide number of tasks. Once a model has been fine-tuned, you won't need to provide examples in the prompt anymore. This saves costs and enables lower-latency requests.
Fine tuning is not available for Chat GPT APIs
https://platform.openai.com/docs/guides/embeddings/what-are-embeddings
Potentially useful for building features like improved versions of search
, related conversations
, automatic labelling
etc
https://platform.openai.com/docs/guides/speech-to-text
The openAI API endpoint is relatively straightforward to use with plain HTTP requests of the following format ref: https://platform.openai.com/docs/api-reference/making-requests
curl https://api.openai.com/v1/completions \
-H "Content-Type: application/json" \
-H "Authorization: Bearer YOUR_API_KEY" \
-d '{"model": "text-davinci-003", "prompt": "Say this is a test", "temperature": 0, "max_tokens": 7}'
Where the key parameters are the API_KEY
for authorization and the prompt
parameter, which dictates how the model responds to various queries.
There are also various libraries available for different language bindings.
I think the OP is a bit quick bu
you tell me if it's too quick but BotPress already did it https://community.botpress.com/t/chatgpt-x-botpress-join-our-webinar-on-march-21st/301
Could you do this using Zapier and a webhook? I have seen something similar done with High Level SMS. If you figure it out, let me know!
what is the best way to use chatgpt on chatwoot?
what is the best way to use chatgpt on chatwoot?
Currently you can create a agent bot connect the user to the GPT models. May be pre-fine-tuning is needed for your Business of customer's business.
what is the best way to use chatgpt on chatwoot?
Currently you can create a agent bot connect the user to the GPT models. May be pre-fine-tuning is needed for your Business of customer's business.
thank you very much, where i find him ? im searching in chatwoot website but i dont find anything related, thanks forr help me
what is the best way to use chatgpt on chatwoot?
Currently you can create a agent bot connect the user to the GPT models. May be pre-fine-tuning is needed for your Business of customer's business.
thank you
I think the Reply Suggestions feature would be even more useful if it were generated based on information from internal knowledge bases.
While one possible solution could be to use Open AI Plugins, unfortunately this feature is currently not available through the API.
Another potential solution would be to build a custom language model application based on frameworks like LangChain and use retrieval augmented generation techniques to achieve this functionality. This would allow the application to retrieve information from the knowledge base and generate a relevant response to the user.
Refs: https://platform.openai.com/docs/plugins/introduction https://docs.langchain.com/docs/use-cases/qa-docs
Update: We are working on a text transformation
feature at the moment as the first step for deeper integration.
ref: Rephrase Agent response. [Generate different copies, Change the tone of text into helpful, professional etc.], https://github.com/chatwoot/chatwoot/pull/6957
Already starting to see some other chats building this.
Eg you create an agent bot, feed it store information eg CMS, product info, blog posts and knowledge base, and it then answers questions about products, shipping, where is my order, and so forth on it's own using the information it has access to. If it can't answer them can then forward to live agent, or live agent can just take over, or contact form if no agent online.
@sojan-official - Can I ask the plans for ChatGPT integration with Chatwoot?
Spec
Exploring ideas for some of the potential implementations of these language models/APIs into Chatwoot.The OpenAI language models provide the following capabilities.
Ref: https://platform.openai.com/docs/introduction/next-steps https://platform.openai.com/docs/models/overview
- Text completion
- Chat completion
- Fine-tuning
- Embeddings
- Speech to text
We could explore building numerous product features leveraging these capabilities.
Text Completion
ref: https://platform.openai.com/docs/guides/completion/inserting-text
These APIs provide us with the option to build various capabilities, sentiment analysis, translations, summarisation etc.
Potential features
- [ ]
Reply Suggestions
Auto-populated reply suggestions for agents generated via the Conversation context, which can be sent by the agent within a couple of clicks. Examples: Gmail / Linkedin reply suggestions- [ ]
Autocomplete Suggestions
Autocomplete suggestions while the agent is typing out the reply based on conversation context. examples: Gmail- [ ]
Summarization
Summarise the large conversation into a gist so that agents can educate themselves about the conversation context in a single glance. Examples: Google Chat conversation summary- [ ]
Transformation
Rephrase Conversations [Write different copies, Change the tone of text into helpful, professional etc.], Translate the text between languages etc.- [ ]
Bots
Bot support agents powered by ChatGPT/GPT3 APIs.- [ ]
Text Expansion
Write short notes or bullet points which will be expanded into full-fledged customer replies or support Center Articles. Examples: Notion AI- [ ]
Sentiment Analysis
System generated CSAT Ratings for conversations where the users haven't provided a feedback rating. This will help the administrators to improve the quality of support provided.Chat completions
ref: https://platform.openai.com/docs/guides/chat
Suited for Agent Bots
OpenAI now has a very specific model,
gpt-3.5-turbo
which allows exploring the above-mentioned use cases with a Chat style API interface. examples use cases suggested by Open AI
- Draft an email or other piece of writing
- Write Python code
- Answer questions about a set of documents
- [ ] Create conversational agents : ( Create Bots in Chatwoot )
- [ ] Give your software a natural language interface : ( Enhance capabilities of our command + k interface )
- Tutor in a range of subjects
- [ ] Translate languages
- Simulate characters for video games and much more
This model is very cost-effective, but at the moment, it doesn't support
fine-tuning
.Fine-tuning
ref: https://platform.openai.com/docs/guides/fine-tuning
GPT-3 has been pre-trained on a vast amount of text from the open internet. When given a prompt with just a few examples, it can often intuit what task you are trying to perform and generate a plausible completion. This is often called "few-shot learning."
Fine-tuning improves on few-shot learning by training on many more examples than can fit in the prompt, letting you achieve better results on a wide number of tasks. Once a model has been fine-tuned, you won't need to provide examples in the prompt anymore. This saves costs and enables lower-latency requests.
- These APIs will allow us to fine-tune the model with data specific to the customer so that we can provide more accurate AI responses.
Fine tuning is not available for Chat GPT APIs
Embedding
https://platform.openai.com/docs/guides/embeddings/what-are-embeddings
Potentially useful for building features like improved versions of
search
,related conversations
,automatic labelling
etcSpeech to text
https://platform.openai.com/docs/guides/speech-to-text
- [ ] Generate transcriptions for audio messages in Chatwoot so that agents can quickly glance at the summary without listening to the whole audio.
- [ ] Will be helpful for features like conversation summarization in case of conversations with audio messages
Using the API
The openAI API endpoint is relatively straightforward to use with plain HTTP requests of the following format ref: https://platform.openai.com/docs/api-reference/making-requests
curl https://api.openai.com/v1/completions \ -H "Content-Type: application/json" \ -H "Authorization: Bearer YOUR_API_KEY" \ -d '{"model": "text-davinci-003", "prompt": "Say this is a test", "temperature": 0, "max_tokens": 7}'
Where the key parameters are the
API_KEY
for authorization and theprompt
parameter, which dictates how the model responds to various queries.There are also various libraries available for different language bindings.
Next Steps
- [ ] Add a GPT Integration to the integration page so that customers can configure their OpenAI API keys into Chatwoot
- [ ] Identify the UI/UX for these features. A lot the features in text completion could be baked into our editor component.
- [ ] Pick out specific use cases into separate issues, prototype / build the features
Update: We have added a reply suggestion and conversation summary option as well. These features will be available on chatwoot cloud already and will be available in the v2.17.0 release.
ref: https://github.com/chatwoot/chatwoot/pull/7029
Here is a quick preview of the functionality:
https://github.com/chatwoot/chatwoot/assets/73185/5a463be2-b5bb-4a98-bc35-b4614511d927
Great feature! But there is a issue with usage besides English Language. What is happening is that some replies and the summarize function is returning always in english, besides the conversation language. What I suggest is that in prompt you should include the commando to output in X language, being X the language set by the admin. That will fix all issues with usage worldwide. That command should be inserted in all prompts.
That issue is happening because the prompt is being made in code in english and usually the OpenAi API will return in the prompt language, or the language sent and command in prompt.
Great feature! But there is a issue with usage besides English Language. What is happening is that some replies and the summarize function is returning always in english, besides the conversation language. What I suggest is that in prompt you should include the commando to output in X language, being X the language set by the admin. That will fix all issues with usage worldwide. That command should be inserted in all prompts.
That issue is happening because the prompt is being made in code in english and usually the OpenAi API will return in the prompt language, or the language sent and command in prompt.
will be fixed in : https://github.com/chatwoot/chatwoot/pull/7092
ref: https://github.com/chatwoot/chatwoot/pull/7162
Generate embedding on support articles to be passed as context for GPT-based Chatbot
Closing this issue since we have the integration in place. We will take further enhancements for the integration in subsequent issues.
This issue has been automatically locked since there has not been any recent activity after it was closed. Please open a new issue for related bugs.
integrate the conversational AI platform into Chatwoot
CW-1551